Banks, airlines and retailers have sold insurance to their customers for decades. Embedded distribution is not a new idea — it is the old idea with three specific things fixed.
Every few years the insurance industry announces a new distribution revolution, and every few years someone old enough points out that banks were selling credit-life cover with loans in the 1970s, airlines were selling travel insurance at the ticket counter before the internet existed, and electronics retailers built entire profit lines on extended warranties. They are right. Embedded insurance is not a new idea. It is affinity distribution — insurance sold through a brand the customer already deals with — rebuilt on different plumbing.
The honest question is therefore not whether the idea is new, but what actually changed. Our answer: three things changed materially, one thing changed partially, and two important things did not change at all. Platforms deciding whether to take embedded distribution seriously should understand all six.
What affinity distribution got right the first time
Classic affinity programmes understood the core insight perfectly: distribution rides on existing trust and existing transactions. A bank offering cover to its mortgage customers does not need to build an insurance brand; the mortgage is the brand. Group schemes for professional associations, cover sold with package holidays, warranties at the till — all of these monetised a relationship someone else had already paid to build.
What they got wrong is also well documented. Products were often generic, priced against no visible alternative, and sold by staff with sales targets rather than product knowledge. Some affinity categories became bywords for poor value, and regulators in several markets eventually intervened in the worst of them. The lesson was not that point-of-need distribution fails. It was that opaque point-of-need distribution eventually gets found out.
Change one: the transaction now carries the data
An airline check-in agent in 1995 knew your destination and travel dates. A digital checkout in 2026 knows the trip, the traveller, the price paid, the device used, and — with consent — enough verified identity to issue a policy without asking a single additional question. The affinity model always had the moment; it rarely had the data to price that moment precisely. Quotes were tiered and approximate because they had to be. Embedded quotes can be exact because the transaction itself is the application form.
This is the deepest change, and it compounds. Precise data means fewer questions; fewer questions mean less drop-off; less drop-off makes thinner-margin products viable at the point of sale, which widens what can be embedded at all.
Change two: the offer became accountable
Affinity insurance was sold, in the literal sense — by a person, with a script, at a moment the customer could not easily compare or decline gracefully. Embedded insurance is presented, inside an interface, where every impression, view, decline and purchase is logged. That sounds like a technical detail. It is actually a governance revolution. A platform can now see exactly what percentage of customers accept an offer, how that changes when the price or copy changes, and whether the product is being taken by people it suits. Regulators can ask for the same evidence. Mis-selling in the old affinity world was discovered years later through complaints; in an instrumented checkout, a product nobody wants reveals itself in the funnel data within weeks.
The old model sold insurance to a captive audience. The new model makes an offer to an audience that can decline in one tap — and records what happened either way.
Change three: the middle layer became infrastructure, not paperwork
Affinity deals were bilateral and bespoke: one insurer, one brand, a negotiated scheme, a binder, and a stack of manual processes holding it together. Setting one up took quarters, which meant only large partners were worth the effort. The infrastructure layer that defines the embedded era — licensed orchestration platforms exposing insurance through APIs — turns that bespoke deal into a repeatable integration. The consequence is not just speed; it is who gets to participate. A mid-sized marketplace that could never have justified a bespoke insurer scheme can now run the same distribution model as a bank.
What changed only partially: the products
Motor, travel, medical, device cover — the categories embedding well today are largely the categories affinity distribution sold decades ago, because they share the same property: standard products where the purchase context supplies the underwriting facts. Product innovation is real but slower than distribution innovation. Anyone claiming the product side has been transformed is ahead of the evidence.
What did not change at all
Two things, and they are the two that decide outcomes.
- Risk still sits with a licensed insurer. Affinity brands were never insurers, and neither are platforms today. Underwriting, solvency and claims liability did not move; only the shop window did.
- Trust is still borrowed, and still repayable. An affinity brand that sold poor cover damaged itself, not just the insurer. That transfer of reputational risk is fully intact in the embedded model — a platform lends its checkout, and takes back whatever experience the customer has at claim time.
Why the distinction matters
If embedded insurance were genuinely new, platforms could be forgiven for treating it as an experiment. Because it is a repaired version of a fifty-year-old model, the standard is higher: the failure modes are documented, and the fixes — precise data, instrumented offers, licensed infrastructure — exist specifically to avoid them. The platforms that win with embedded distribution will be the ones that treat it as the affinity model with the excuses removed.